Token导航 LogoToken导航

Agent Skill · 研究检索

commit(提交)

commit 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

commit 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为需要联网;许可证信息还不明确,第三方扫描已通过,按安装量在研究检索类的 73,675 个 Skill 里排第 49,707。

总安装 642周安装 27GitHub Stars 公开资料未说明下载 225核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
unknown

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:commit(提交)
来源仓库:https://github.com/kelvinz/cobb
仓库路径:skills/commit
安装命令:
npx skills add https://github.com/kelvinz/cobb --skill commit
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/kelvinz/cobb --skill commit

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

commit 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中快速定位候选结果。适用于根据关键词、任务场景或来源线索进行信息定位和整理。通过 npx skills add 命令从指定 GitHub 仓库安装使用。

安装前需确认权限范围、维护状态及是否涉及联网或文件操作。commit 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

来源仓库内路径:skills/commit

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/kelvinz/cobb --skill commit;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要需要联网,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

commit 怎么安装?

在终端执行 npx skills add https://github.com/kelvinz/cobb --skill commit。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

commit 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

commit 需要什么权限?

权限级别是需要联网。该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

commit 是免费的吗?

公开资料没有标注明确许可证,收费和商用限制都无法确认。正式使用前请到来源仓库核对授权说明。

commit 用的人多吗?

公开数据是累计安装 642 次,近一周安装 27 次,下载 225 次。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

commit 安全吗?

第三方扫描结果为通过,但这只覆盖已知的静态检查项,不能替代自己读一遍源码和安装命令。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。
  • 第三方扫描已通过,常见的高风险行为没有命中。

先想清楚的地方

  • 许可证没有明确标注,能不能商用、能不能再分发都不确定。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。

不太适合

  • 需要明确授权才能用到公司项目里的场景。
  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
commit

提交 · 本页

642—需要联网未明确通过

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

commit

Commit changes in atomic steps, then finalise and clean up the feature branch when requested.


Guardrails

  • Require user confirmation before every commit.
  • For every atomic commit proposal, provide a proposed title.
  • Provide a proposed body before asking for approval.
  • When asking for a single user decision, provide numbered short-reply options (for example: 1, 2, 3, 4).
  • For bundled finalise decisions, use field+choice codes (for example: 1A, 2B, 3A) so users can reply in one line.
  • For bundled finalise decisions, also offer 0 (default) as a short reply. Treat 0 and default as 1A 2B 3B 4A 5B.
  • Keep commits atomic; if a title needs "and", split the change set.
  • Never mix unrelated files in one commit.
  • Determine commit type from the actual diff intent, not from branch name, file paths, or habit.
  • Never use chore for behaviour changes (flow, API/data semantics, bug fixes, or security fixes).
  • If a change group contains multiple intents, split it before proposing.
  • Examples: feat + fix, fix + chore.
  • Never add AI attribution or Co-authored-by trailers unless the user explicitly asks.
  • Do not push to remote without explicit user confirmation.
  • In finalise mode, collect merge target, merge strategy, push choice, and local/remote branch deletion choices in one decision prompt.
  • Treat explicit yes/no decisions from that prompt as confirmation; only ask follow-ups for missing or conflicting inputs.
  • Do not assume main; it may be dev or another branch.
  • Before final merge, sync with the target branch and resolve any conflicts.
  • Use review as a quality gate before commit/finalise when needed.
  • Capture context inline: update tasks/context.md during commit/finalise when durable info appears.
  • Do not add a separate context-only step.
  • Keep PRD checklist updates inside the atomic commit that completes that work.
  • Never make a trailing commit only to catch up PRD checklist or context updates.
  • In finalise, bundle tracking updates into one pre-merge commit.
  • Include PRD archive and context updates as needed.
  • If no tracking changes are needed, skip the finalise commit and state why.
  • Use finalise tracking for closeout only.
  • Do not retroactively log completed atomic work unless the user explicitly asks.
  • Follow the merge strategy resolution order in references/finalise-policy.md.
  • Never delete the base/default branch.
  • Never delete the currently checked-out branch.
  • Require explicit user confirmation before deleting local or remote branches.
  • Do not claim tests passed or checks succeeded without actually running them; if you didn't run it, say so.

Commit Message Rules

Title format (required)

<emoji> <type>: <imperative summary>

  • type must be one of: feat, fix, chore
  • Summary must be short, specific, and imperative (e.g., "add", "fix", "remove", "refactor")
  • Avoid vague summaries like "update code" or "fix stuff"

Finalise title (required)

For a finalise commit, require:

🧹 chore: finalise f-## <short-summary>

Keep the summary short and imperative.

Emoji mapping (default)

  • feat → ✨
  • fix → 🐛
  • chore → 🧹

Use a different emoji only if it more precisely matches the change.

Type and body guidance (required)

Use the detailed rubric and templates in references/commit-rules.md for:

  • selecting feat vs fix vs chore
  • classifying mixed-intent diffs
  • writing standard commit bodies
  • writing finalise commit bodies

Read this reference before proposing the first commit in a session, and revisit it whenever classification is ambiguous.


Modes

  • commit mode: propose and execute one atomic commit at a time.
  • finalise mode: merge/close the completed branch and delete feature branches.
  • hotfix mode: for urgent fixes committed directly to the default branch.
  • Requires review approval and a tasks/context.md hotfix rationale update.

Workflow

  1. Inspect current changes (git status --short, git diff, git diff --staged). Identify the active feature PRD when applicable.
  2. Partition changes into atomic commit groups and map each group to:

- PRD checklist/user-story items completed by that group (if any) - context-worthy outcomes produced by that group (if any)

  1. For the next group, propose:

- files/hunks included - type rationale (why this is feat / fix / chore, citing concrete diff intent) - PRD checklist lines to update in this same commit (or explicit none with reason) - context updates to include in this same commit (or explicit none with reason) - commit title (emoji + type + imperative summary) - commit body (Summary/Why/Context/Alternatives/Trade-offs/Consequences)

  1. Ask for a decision (accept number or keyword):

- 1 (yes): stage only that group and commit - 2 (edit): revise message/scope and ask again - 3 (skip): leave group uncommitted and move on - 4 (split): break the group into smaller commits and repropose

  1. After each accepted commit:

- commit immediately - report commit hash + title + short "what changed" summary - propose the next atomic group until done

  1. During commit mode, ensure tracking is coupled to the atomic change:

- include PRD checklist updates for completed stories in that same commit - include context updates only when durable rationale/gotchas emerge from that same commit - if no PRD/context updates are needed for a group, explicitly mark them as none with reason


Finalise Mode

Use after all intended commits are done.

  1. Confirm preconditions:

- working tree clean - on feature branch (not base)

  1. Determine the feature PRD path to finalise:

- prefer active feature PRD files in tasks/ matching the feature ID (tasks/f-##-*.md) - if multiple matches exist or feature ID is unclear, ask the user for the exact path

  1. Before merge, prepare tracking updates:

- if the PRD path matches tasks/f-##-<slug>.md, ensure tasks/archive/ exists - move the PRD to tasks/archive/f-##-<slug>.md (same filename); if already archived, skip - if tasks/context.md exists (or should exist), update it when durable state changed: - add completed milestone entry for the finalised feature - update "Current state" (next up / blockers) if it changed - capture any key decisions or gotchas discovered during completion - stage all resulting tracking changes together - propose exactly one 🧹 chore: finalise f-##... commit (use short finalise body) - ask for user confirmation before committing (same 1/2/3 or yes/edit/skip options as commit mode) - if no tracking changes are needed, explicitly state why and skip the finalise commit - do not use this finalise commit to catch up missed atomic PRD/context updates - only do that if the user explicitly approves

  1. Collect one finalise decision bundle before merge/cleanup:

- ask once for: target branch, merge strategy, push-after-merge, delete-local-branch, delete-remote-branch - merge strategy choices: auto (policy-resolved), merge-commit, linear-history, squash, rebase - include a default shortcut line: 0 / default = 1A 2B 3B 4A 5B - present options with field+choice codes: - 1A/1B/... for target branch choices; make 1A the resolved default target branch and include 1X for custom branch text - 2A=auto, 2B=merge-commit, 2C=linear-history, 2D=squash, 2E=rebase - 3A=yes push-after-merge, 3B=no push-after-merge - 4A=yes delete-local-branch, 4B=no delete-local-branch - 5A=yes delete-remote-branch, 5B=no delete-remote-branch - accept compact replies like 1B 2A 3A 4B 5B, 0, or default, plus keyword/freeform equivalents - if any field is missing or ambiguous, ask only for the missing field(s) - treat explicit yes/no values in this bundle as the required confirmations for push and branch deletion

  1. Confirm finalise gate before merging:

- require review: Good to commit: Yes; if missing, run review before merging - ensure the branch is synced with the target; if not, sync and rerun review

  1. Resolve the merge strategy per references/finalise-policy.md.

- if bundle strategy is auto, present the resolved strategy as a suggestion and let the user confirm or override - if bundle strategy is explicit, use it directly unless policy/safety checks require an override

  1. Merge feature branch into the target branch using the confirmed strategy.
  2. If push-after-merge is yes, push target branch to remote; if no, skip and state that remote update is pending.
  3. Apply branch safety checks before deletion:

- confirm <feature-branch> is not the target branch - confirm <feature-branch> is not the default branch - confirm current HEAD is not <feature-branch> when deleting it

  1. Delete completed feature branch according to bundle confirmations:
  • if delete-local-branch is yes: git branch -d <feature-branch>; otherwise skip
  • if delete-remote-branch is yes: git push origin --delete <feature-branch>; otherwise skip

Hotfix Mode

Use for urgent fixes committed directly to the default branch without a feature branch.

  1. Confirm on the default branch (not a feature branch).
  2. Require review approval before committing.
  3. Follow the standard atomic commit workflow (steps 1-6 above).

- PRD checklist: mark as none — hotfixes typically have no PRD. - Type: use fix unless the change is purely non-behavioural (chore).

  1. Record hotfix rationale in tasks/context.md (what broke, why the hotfix was necessary, follow-up actions).
  2. No finalise step needed — already on the default branch.

Output

For each proposed commit, provide:

  • atomic scope (files/hunks)
  • short "what changed" summary
  • type rationale (1-2 lines; evidence-based)
  • proposed title
  • proposed body
  • explicit confirmation prompt (Commit this change now?)

After execution, provide:

  • created commit hash/title
  • short "what changed" summary
  • remaining uncommitted groups
  • finalise recommendation when all groups are complete
  • PRD checklist sync status (updated or none with reason)
  • context sync status (updated or skipped with reason)
  • End with a short status block:

- Files changed: list of created/updated files - Key decisions: any assumptions or choices made (if any) - Next step: recommended next skill or action

平台分布

Codex

37.27%
按下载量换算84

Claude

27.54%
按下载量换算62

Cursor

17.99%
按下载量换算40

Gemini CLI

8.35%
按下载量换算19

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills